zgemm_kernel_2x2_sse.s

来自「Optimized GotoBLAS libraries」· S 代码 · 共 1,540 行 · 第 1/3 页

S
1,540 行
字号
123
	movaps	%xmm1,  4 * SIZE(BB)	movaps	%xmm2,  8 * SIZE(BB)	movaps	%xmm3, 12 * SIZE(BB)	addl	$ 4 * SIZE, %edi	ALIGN_4.L05:	movl	C,  %esi	movl	A,  %edx	movl	M,  %ebx	sarl	$1, %ebx	jle	.L30	ALIGN_4.L10:#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	leal	BUFFER, BB	# boffset1 = boffset#else	leal	BUFFER, BB	# boffset1 = boffset	movl	KK, %eax	leal	(, %eax,   8), %eax	leal	(AA, %eax, 2), AA	leal	(BB, %eax, 8), BB#endif		movaps	 0 * SIZE(AA), %xmm0	XORPS	%xmm4, %xmm4	movaps	16 * SIZE(AA), %xmm1	XORPS	%xmm5, %xmm5	movaps	 0 * SIZE(BB), %xmm2	XORPS	%xmm6, %xmm6	movaps	16 * SIZE(BB), %xmm3	XORPS	%xmm7, %xmm7#if defined(OPTERON) || defined(BARCELONA)	prefetchw   4 * SIZE(%esi)	prefetchw   4 * SIZE(%esi, LDC)#endif#ifdef PENTIUM4	prefetchnta 4 * SIZE(%esi)	prefetchnta 4 * SIZE(%esi, LDC)#endif#ifndef TRMMKERNEL	movl	K, %eax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movl	K, %eax	subl	KK, %eax	movl	%eax, KKK	#else	movl	KK, %eax#ifdef LEFT	addl	$2, %eax#else	addl	$2, %eax#endif	movl	%eax, KKK#endif#if 1	andl	$-8, %eax	sall	$4, %eax	je	.L15.L1X:		KERNEL1(32  *  0)	KERNEL2(32  *  0)	KERNEL3(32  *  0)	KERNEL4(32  *  0)	KERNEL5(32  *  0)	KERNEL6(32  *  0)	KERNEL7(32  *  0)	KERNEL8(32  *  0)	cmpl	$128 *  1, %eax	jle	.L12	KERNEL1(32  *  1)	KERNEL2(32  *  1)	KERNEL3(32  *  1)	KERNEL4(32  *  1)	KERNEL5(32  *  1)	KERNEL6(32  *  1)	KERNEL7(32  *  1)	KERNEL8(32  *  1)	cmpl	$128 *  2, %eax	jle	.L12	KERNEL1(32  *  2)	KERNEL2(32  *  2)	KERNEL3(32  *  2)	KERNEL4(32  *  2)	KERNEL5(32  *  2)	KERNEL6(32  *  2)	KERNEL7(32  *  2)	KERNEL8(32  *  2)	cmpl	$128 *  3, %eax	jle	.L12	KERNEL1(32  *  3)	KERNEL2(32  *  3)	KERNEL3(32  *  3)	KERNEL4(32  *  3)	KERNEL5(32  *  3)	KERNEL6(32  *  3)	KERNEL7(32  *  3)	KERNEL8(32  *  3)	cmpl	$128 *  4, %eax	jle	.L12	KERNEL1(32  *  4)	KERNEL2(32  *  4)	KERNEL3(32  *  4)	KERNEL4(32  *  4)	KERNEL5(32  *  4)	KERNEL6(32  *  4)	KERNEL7(32  *  4)	KERNEL8(32  *  4)	cmpl	$128 *  5, %eax	jle	.L12	KERNEL1(32  *  5)	KERNEL2(32  *  5)	KERNEL3(32  *  5)	KERNEL4(32  *  5)	KERNEL5(32  *  5)	KERNEL6(32  *  5)	KERNEL7(32  *  5)	KERNEL8(32  *  5)	cmpl	$128 *  6, %eax	jle	.L12	KERNEL1(32  *  6)	KERNEL2(32  *  6)	KERNEL3(32  *  6)	KERNEL4(32  *  6)	KERNEL5(32  *  6)	KERNEL6(32  *  6)	KERNEL7(32  *  6)	KERNEL8(32  *  6)	cmpl	$128 *  7, %eax	jle	.L12	KERNEL1(32  *  7)	KERNEL2(32  *  7)	KERNEL3(32  *  7)	KERNEL4(32  *  7)	KERNEL5(32  *  7)	KERNEL6(32  *  7)	KERNEL7(32  *  7)	KERNEL8(32  *  7)	addl	$128 * 8  * SIZE, BB	addl	$128 * 2  * SIZE, AA	subl	$128 * 8, %eax	jg	.L1X	jmp	.L15.L12:	leal	(AA, %eax, 1), AA	leal	(BB, %eax, 4), BB	ALIGN_4#else	sarl	$3, %eax	je	.L15	ALIGN_4.L11:	KERNEL1(32  *  0)	KERNEL2(32  *  0)	KERNEL3(32  *  0)	KERNEL4(32  *  0)	KERNEL5(32  *  0)	KERNEL6(32  *  0)	KERNEL7(32  *  0)	KERNEL8(32  *  0)	addl	$ 32 * SIZE, AA	addl	$128 * SIZE, BB	decl   %eax	jne    .L11	ALIGN_4#endif	.L15:#ifndef TRMMKERNEL	movl	K, %eax#else	movl	KKK, %eax#endif	movaps	ALPHA_R, %xmm1	movaps	ALPHA_I, %xmm3	andl	$7, %eax		# if (k & 1)	BRANCH	je .L14	ALIGN_4.L13:	mulps	%xmm0, %xmm2	addps	%xmm2, %xmm4	movaps	 4 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	addps	%xmm2, %xmm5	movaps	 8 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	mulps	12 * SIZE(BB), %xmm0	addps	%xmm2, %xmm6	movaps	16 * SIZE(BB), %xmm2	addps	%xmm0, %xmm7	movaps	 4 * SIZE(AA), %xmm0	addl	$ 4 * SIZE, AA	addl	$16 * SIZE, BB	decl	%eax	jg	.L13	ALIGN_4.L14:	shufps	$0xb1, %xmm5, %xmm5	shufps	$0xb1, %xmm7, %xmm7#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \    defined(RR) || defined(RC) || defined(CR) || defined(CC)	subps	%xmm5, %xmm4	subps	%xmm7, %xmm6#else	addps	%xmm5, %xmm4	addps	%xmm7, %xmm6#endif	movaps	%xmm4, %xmm5	movaps	%xmm6, %xmm7	shufps	$0xb1, %xmm4, %xmm4	shufps	$0xb1, %xmm6, %xmm6	mulps	%xmm1, %xmm5	mulps	%xmm3, %xmm4	mulps	%xmm1, %xmm7	mulps	%xmm3, %xmm6	addps	%xmm5, %xmm4	addps	%xmm7, %xmm6#ifndef TRMMKERNEL	shufps	$0xe4, %xmm0, %xmm0	movsd	0 * SIZE(%esi), %xmm0	movhps	2 * SIZE(%esi), %xmm0	shufps	$0xe4, %xmm2, %xmm2	movsd	0 * SIZE(%esi, LDC), %xmm2	movhps	2 * SIZE(%esi, LDC), %xmm2	addps	%xmm0, %xmm4	addps	%xmm2, %xmm6#endif	movsd	%xmm4, 0 * SIZE(%esi)	movhps	%xmm4, 2 * SIZE(%esi)	movsd	%xmm6, 0 * SIZE(%esi, LDC)	movhps	%xmm6, 2 * SIZE(%esi, LDC)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movl	K, %eax	subl	KKK, %eax	leal	(,%eax, 8), %eax	leal	(AA, %eax, 2), AA	leal	(BB, %eax, 8), BB#endif#if defined(TRMMKERNEL) && defined(LEFT)	addl	$2, KK#endif	addl	$4 * SIZE, %esi		# coffset += 4	decl	%ebx			# i --	jg	.L10	ALIGN_4.L30:	movl	M,  %ebx	andl	$1, %ebx	jle	.L99	ALIGN_4.L40:#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	leal	BUFFER, BB	# boffset1 = boffset#else	leal	BUFFER, BB	# boffset1 = boffset	movl	KK, %eax	leal	(, %eax,    8), %eax	leal	(AA, %eax,  1), AA	leal	(BB,  %eax, 8), BB#endif		XORPS	%xmm4, %xmm4	XORPS	%xmm5, %xmm5	XORPS	%xmm6, %xmm6	XORPS	%xmm7, %xmm7	movlps	 0 * SIZE(AA), %xmm0	movlps	 8 * SIZE(AA), %xmm1	movaps	 0 * SIZE(BB), %xmm2	movaps	16 * SIZE(BB), %xmm3#ifndef TRMMKERNEL	movl	K, %eax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movl	K, %eax	subl	KK, %eax	movl	%eax, KKK	#else	movl	KK, %eax#ifdef LEFT	addl	$1, %eax#else	addl	$2, %eax#endif	movl	%eax, KKK#endif	sarl	$3, %eax	je	.L42	ALIGN_4.L41:	mulps	%xmm0, %xmm2	prefetcht1	(PREFETCHSIZE +  0) * SIZE(AA)	addps	%xmm2, %xmm4	movaps	 4 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	addps	%xmm2, %xmm5	movaps	 8 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	mulps	12 * SIZE(BB), %xmm0	addps	%xmm2, %xmm6	movaps	32 * SIZE(BB), %xmm2	addps	%xmm0, %xmm7	movlps	 2 * SIZE(AA), %xmm0	mulps	%xmm0, %xmm3	addps	%xmm3, %xmm4	movaps	20 * SIZE(BB), %xmm3	mulps	%xmm0, %xmm3	addps	%xmm3, %xmm5	movaps	24 * SIZE(BB), %xmm3	mulps	%xmm0, %xmm3	mulps	28 * SIZE(BB), %xmm0	addps	%xmm3, %xmm6	movaps	48 * SIZE(BB), %xmm3	addps	%xmm0, %xmm7	movlps	 4 * SIZE(AA), %xmm0	mulps	%xmm0, %xmm2	addps	%xmm2, %xmm4	movaps	36 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	addps	%xmm2, %xmm5	movaps	40 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	mulps	44 * SIZE(BB), %xmm0	addps	%xmm2, %xmm6	movaps	64 * SIZE(BB), %xmm2	addps	%xmm0, %xmm7	movlps	 6 * SIZE(AA), %xmm0	mulps	%xmm0, %xmm3	addps	%xmm3, %xmm4	movaps	52 * SIZE(BB), %xmm3	mulps	%xmm0, %xmm3	addps	%xmm3, %xmm5	movaps	56 * SIZE(BB), %xmm3	mulps	%xmm0, %xmm3	mulps	60 * SIZE(BB), %xmm0	addps	%xmm3, %xmm6	movaps	80 * SIZE(BB), %xmm3	addps	%xmm0, %xmm7	movlps	16 * SIZE(AA), %xmm0	mulps	%xmm1, %xmm2#if defined(OPTERON) || defined(BARCELONA)	prefetcht1     (PREFETCHSIZE + 16) * SIZE(AA)#endif	addps	%xmm2, %xmm4	movaps	68 * SIZE(BB), %xmm2	mulps	%xmm1, %xmm2	addps	%xmm2, %xmm5	movaps	72 * SIZE(BB), %xmm2	mulps	%xmm1, %xmm2	mulps	76 * SIZE(BB), %xmm1	addps	%xmm2, %xmm6	movaps	96 * SIZE(BB), %xmm2	addps	%xmm1, %xmm7	movlps	10 * SIZE(AA), %xmm1	mulps	%xmm1, %xmm3	addps	%xmm3, %xmm4	movaps	84 * SIZE(BB), %xmm3	mulps	%xmm1, %xmm3	addps	%xmm3, %xmm5	movaps	88 * SIZE(BB), %xmm3	mulps	%xmm1, %xmm3	mulps	92 * SIZE(BB), %xmm1	addps	%xmm3, %xmm6	movaps	112 * SIZE(BB), %xmm3	addps	%xmm1, %xmm7	movlps	12 * SIZE(AA), %xmm1	mulps	%xmm1, %xmm2	addps	%xmm2, %xmm4	movaps	100 * SIZE(BB), %xmm2	mulps	%xmm1, %xmm2	addps	%xmm2, %xmm5	movaps	104 * SIZE(BB), %xmm2	mulps	%xmm1, %xmm2	mulps	108 * SIZE(BB), %xmm1	addps	%xmm2, %xmm6	movaps	128 * SIZE(BB), %xmm2	addps	%xmm1, %xmm7	movlps	14 * SIZE(AA), %xmm1	mulps	%xmm1, %xmm3	addps	%xmm3, %xmm4	movaps	116 * SIZE(BB), %xmm3	mulps	%xmm1, %xmm3	addps	%xmm3, %xmm5	movaps	120 * SIZE(BB), %xmm3	mulps	%xmm1, %xmm3	mulps	124 * SIZE(BB), %xmm1	addps	%xmm3, %xmm6	movaps	144 * SIZE(BB), %xmm3	addps	%xmm1, %xmm7	movlps	 24 * SIZE(AA), %xmm1	addl	$ 16 * SIZE, AA	addl	$128 * SIZE, BB	decl	%eax	jne	.L41	ALIGN_4	.L42:#ifndef TRMMKERNEL	movl	K, %eax#else	movl	KKK, %eax#endif	movaps	ALPHA_R, %xmm1	movaps	ALPHA_I, %xmm3	andl	$7, %eax		# if (k & 1)	BRANCH	je .L44	ALIGN_4.L43:	mulps	%xmm0, %xmm2	addps	%xmm2, %xmm4	movaps	 4 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	addps	%xmm2, %xmm5	movaps	 8 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	mulps	12 * SIZE(BB), %xmm0	addps	%xmm2, %xmm6	movaps	16 * SIZE(BB), %xmm2	addps	%xmm0, %xmm7	movlps	 2 * SIZE(AA), %xmm0	addl	$ 2 * SIZE, AA	addl	$16 * SIZE, BB	decl	%eax	jg	.L43	ALIGN_4.L44:	shufps	$0xb1, %xmm5, %xmm5	shufps	$0xb1, %xmm7, %xmm7#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \    defined(RR) || defined(RC) || defined(CR) || defined(CC)	subps	%xmm5, %xmm4	subps	%xmm7, %xmm6#else	addps	%xmm5, %xmm4	addps	%xmm7, %xmm6#endif	movaps	%xmm4, %xmm5	movaps	%xmm6, %xmm7	shufps	$0xb1, %xmm4, %xmm4	shufps	$0xb1, %xmm6, %xmm6	mulps	%xmm1, %xmm5	mulps	%xmm3, %xmm4	mulps	%xmm1, %xmm7	mulps	%xmm3, %xmm6	addps	%xmm5, %xmm4	addps	%xmm7, %xmm6#ifndef TRMMKERNEL	shufps	$0xe4, %xmm4, %xmm4	shufps	$0xe4, %xmm6, %xmm6	movsd	0 * SIZE(%esi), %xmm0	movsd	0 * SIZE(%esi, LDC), %xmm2	addps	%xmm0, %xmm4	addps	%xmm2, %xmm6#endif	movsd	%xmm4, 0 * SIZE(%esi)	movsd	%xmm6, 0 * SIZE(%esi, LDC)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movl	K, %eax	subl	KKK, %eax	leal	(,%eax, 8), %eax	leal	(AA, %eax, 1), AA	leal	(BB, %eax, 8), BB
123

zgemm_kernel_2x2_sse.s - 源码说明

本页面展示了「Optimized GotoBLAS libraries」中的 zgemm_kernel_2x2_sse.s 源码文件，采用 S 编程语言编写，共 1,540 行代码。您可以在线阅读完整代码内容，也可以返回资源详情页下载完整源码包进行本地学习和开发。

虫虫开发者社区收录了大量与BLAS相关的技术资源，包括源代码、技术文档、电路图等，是电子工程师和嵌入式开发者的专业学习平台。

⌨️ 快捷键说明

复制代码Ctrl + C
搜索代码Ctrl + F
全屏模式F11
增大字号Ctrl + =
减小字号Ctrl + -
显示快捷键?